From Context to Conversation: Building AI-Powered Workflows at Scale

What an electrifying evening at the Artefact offices! 🎉
From sandboxing AI agents to scaling e-commerce bots, the night was a masterclass in GenAI. 🧠✨

Talk 1: Adithya Krishnan - Agents in the Dataverse 🦆
▪️ Agents as users: Provisioning isolated compute (Motherduck / MCP) eliminates resource contention and allows for easy shutdown.
▪️ Data Sandboxing: Zero-copy clones let agents safely explore tables and test SQL without risking production data.
▪️ Context: Push context management down to the database object level to prevent shared context overflow.

Talk 2: Diederik Heijbroek - Fixing Order Picking Errors 🌷
A Royal Flora Holland GenAI (Streamlit, LangGraph) implementation to tackle fragmented warehouse data—now used to onboard new employees!
▪️ Context biases reasoning: Supplying the wrong priority or too much context distracts the LLM from the goal.
▪️ Embrace the 97%: Retrieval is rarely perfect. Build systems where humans review the final judgment.
▪️ Start small: Involve stakeholders from day one and upskill the tech team.

Talk 3: Lorenzo Casimo - Reliable Conversational Agents 💄
A multi-agent architecture for a beauty brand, designed to drive conversions while avoiding hallucinations.
▪️ Modular execution: Split responsibilities into specialized agents (Q&A, Vertex AI Product, Handover).
▪️ Evaluation is critical: Maintaining a strict "Golden test set" for simple QA is non-negotiable.
▪️ Trust > Tech: Hard guardrails are essential. Building human trust remains the hardest part of conversational AI.

Massive thanks to Artefact for hosting! 🙏
🚀 Artefact is Hiring! Looking for 2 Data Scientists / AI Engineers to build advanced AI agents and robust ML solutions.

Reach out to Luam Tesfamarian to learn more!